Universal model for UV/Vis spectroscopy of gold nanoparticles

Machine learning models have limited interoperability in analytical chemistry; mostly.

Experiments always vary between labs. This can be enough to disturb machine learning models trained on small data.

• Different equipment used
• Different sample preparation
• Different calibration methods
• Differences in executions

The list goes on and on.

Here is a beautiful example of a “universal machine learning model” Gleason et al. (2024) for UV/Vis spectroscopy of gold nanoparticles. A method that almost always works!

From spectrum to population from Gleason et al. (2024)

Gold Nanorod shape distribution prediction

The model is trained on simulated UV/Vis spectra of gold nanorods with different aspect ratios, and it can predict the shape distribution from the spectrum.

The method is called AuNR-SMA and outlined in the paper “Automated Gold Nanorod Spectral Morphology Analysis Pipeline”. It is a computational method, but not machine learning.

Computational workflow of AuNR-SMA Gleason et al. (2024)

The raw material is a bag of simulated spectra for gold nanorods (AuNRs) with varying aspect ratios and dimensions. The authors assume that the particles dimensions follow a Gaussian distribution. The observed spectrum is then reconstructed sum of spectra of individual particles. There is no learning involved, just physics and math.

✅ Interoperability Comparison: AuNR-SMA vs Traditional Methods

Feature	AuNR-SMA	Traditional TEM	Empirical UV-Vis Methods
🔬 Equipment Required	✅ Any UV-Vis-NIR spectrophotometer	❌ Expensive TEM facility	⚠️ Specific calibrated instrument
🌍 Cross-Lab Transfer	✅ Physics-based = Universal	✅ Direct imaging	❌ Requires recalibration
⏱️ Analysis Time	✅ <5 minutes	❌ Hours + sample prep	✅ Minutes
💰 Cost per Sample	✅ <$1	❌ $50-200	✅ <$1
📊 Information Obtained	✅ Full size distributions	✅ Individual particles	⚠️ Only average values
🤖 Automation Ready	✅ Fully automated	⚠️ Limited	⚠️ Instrument-specific
👩‍🔬 Expertise Needed	✅ Minimal training	❌ Specialized training	⚠️ Method development
📈 Throughput	✅ 100s/day	❌ 10s/day	✅ 100s/day

🎯 What Makes AuNR-SMA Universally Transferable

Enabler

The paper mentions explores three applications.

1. The authors automate the analysis of one-pot seedless high throughput AuNR synthesis.
2. Then they train machine learning models to predict AuNR synthesis outcomes
3. They use spectra from the literature to infer population level data that was not reported

Results prediciton systehesis distribution with machine learning Gleason et al. (2024)

Any type of lab can benefit from this method.

Lab Type	How AuNR-SMA Helps
🏭 Industrial Scale-Up	Monitor batch-to-batch consistency without TEM delays
🔬 Academic Research	Publish complete size distributions, not just TEM samples
🏥 Biomedical Applications	Rapid QC for therapeutic nanoparticles
🤝 Collaborative Projects	Same analysis method across all partner labs
📖 Literature Review	Extract quantitative data from published spectra

References

Gleason, Samuel P., Jakob C. Dahl, Mahmoud Elzouka, Xingzhi Wang, Dana O. Byrne, Hannah Cho, Mumtaz Gababa, et al. 2024. “Automated Gold Nanorod Spectral Morphology Analysis Pipeline.” ACS Nano 18 (51): 34646–55. https://doi.org/10.1021/acsnano.4c09753.